On Efficient Use of Logistic Regression When a Continuous Biomarker Is Assayed on Pooled Biospecimens

• Activity Number: 183
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Robert Lyles*+ and Emily Mitchell and Clarice Weinberg and David M. Umbach and Enrique Schisterman
• Companies: Emory University and NICHD and NIEHS and NIEHS and NICHD
• Keywords: Efficiency ; Exposure ; Pooling ; Study design

Abstract

Reductions in lab assay costs afforded by the pooling of biospecimens have long been recognized in epidemiological disease surveillance, and more recently have motivated design and analytic developments in regression settings. For example, Weinberg and Umbach (1999) facilitated the fitting of logistic regression models when pooling is used to assess a continuous exposure or biomarker. Our focus is on improving efficiency in this setting by using subject-specific information at the pool allocation stage. Specifically, we promote pooling within strata defined by the outcome and other covariates for the purpose of precision. A consequence of such a design strategy is that the analysis must account for outcome/covariate-dependent pooling to ensure validity in estimated poolwise logistic regression coefficients. We propose important modifications to the offsets suggested by Weinberg and Umbach, and we use simulations and a real-data example to illustrate the benefits of the recommended design measures relative to pooling only within outcome strata. As a side benefit, the newly-defined offsets yield valid estimates of covariate effects in the presence of interactions with exposure.